Speech Command Recognition (SCR) has many applications in smart home systems, voice-controlled robots, and voice assistants. The modern SCR systems employ deep learning models trained on speech command datasets. Nowadays, the Google Speech Commands (GSC) dataset serves as a benchmark for training and evaluating SCR models, but it only includes English commands. In this work, we introduce an equivalent of the GSC for the Kazakh, Tatar, and Russian languages. We collected the dataset via a Telegram bot in real-world settings. The dataset includes the same 35 commands as GSC and contains 3,623 utterances for Kazakh, 3,547 for Tatar, and 1,625 for Russian. We trained and evaluated both monolingual and multilingual SCR models. The experimental results show that the multilingual model outperforms the monolingual models with an accuracy of 98.57%, 99.33%, 96.86%, and 97.33% on the test sets of the Kazakh, Tatar, Russian datasets and GSC, respectively. Furthermore, we deployed the model on an NVIDIA Jetson Orin NX 16GB, achieving an average inference time of 0.008 seconds. This performance makes it suitable for real-time applications. We have made the dataset, source code, and pretrained models publicly available to encourage further research in this area.

This paper introduces a novel audio-to-image encoding framework that integrates multiple dimensions of voice characteristics into a single RGB image for speaker recognition. In this method, the green channel encodes raw audio data, the red channel embeds statistical descriptors of the voice signal (including key metrics such as median and mean values for fundamental frequency, spectral centroid, bandwidth, rolloff, zero-crossing rate, MFCCs, RMS energy, spectral flatness, spectral contrast, chroma, and harmonic-to-noise ratio), and the blue channel comprises subframes representing these features in a spatially organized format. A deep convolutional neural network trained on these composite images achieves 98% accuracy in speaker classification across 2 speakers, suggesting that this integrated multichannel representation can provide a more discriminative input for voice recognition tasks.

The implementation of quantum algorithms is attracting a lot of attention from the research community, mainly due to the technical advances made and the democratization of access to quantum computers. However, it is important for researchers and practitioners to learn from the methodologies of the classical-computing world in order to ensure the completeness, correctness, and validation of the produced software. Arguably, taking the right steps would contribute to the correct progress of this still-young field, aiming not to lose the attraction of the industry. With this motivation in mind, we provide in this paper a brief review of tools currently available to cope with the testing of quantum software. As part of this review, we have also detected possible gaps and opportunities that could be useful for the current community.

Cricket is a dynamic sport requiring strategic decision-making from both batsmen and bowlers. Traditional coaching methods rely on human expertise and statistical analysis, but with advancements in artificial intelligence, reinforcement learning (RL) offers a data-driven approach to optimize cricket strategies. This paper introduces a Deep Q-Network (DQN)-based Cricket Strategy Optimizer, designed to enhance decisionmaking in batting and bowling. The proposed model leverages reinforcement learning to train two AI agents one for batsmen and another for bowlers enabling them to learn optimal actions under various game conditions. For the batting agent, the model predicts the best shot selection based on factors such as bowler type, field placements, and match scenarios. The bowling agent learns to select the most effective deliveries and field setups to maximize wicket-taking potential and minimize runs conceded. The training process involves simulating thousands of game situations, where the AI receives rewards for successful outcomes (e.g., scoring runs efficiently, taking wickets) and penalties for suboptimal decisions (e.g., losing wickets, conceding boundaries). The Deep Q-Network (DQN) model iteratively optimizes its decision-making by modelling the game environment as a Markov Decision Process (MDP), where each state-action transition is refined through continuous learning. Extensive simulations demonstrate that the trained AI agents can outperform conventional heuristics and adapt dynamically to different game contexts. This study establishes a foundation for AI-powered cricket analytics, offering potential applications in player training, strategy formulation, and match simulations. Future work will explore multi-agent reinforcement learning, and integration with real match data for further validation.

In this paper, the problem of intra-operative hypotension (IOH) prediction is addressed. Recent studies on the Hypotension Prediction Index have demonstrated a gap between the results presented during model development and clinical evaluation. Thus, there is a need for better collaboration between data scientists and clinicians, who need to agree on a common basis. In this contribution, we propose a comprehensive framework for IOH prediction: to address several issues inherent to the commonly used fixed-time-to-onset approach in the literature, a sliding window approach is suggested. The risk prediction problem is formalized with consistent precision-recall metrics rather than the receiver-operator characteristic. For illustration, a standard machine learning method is applied using two different datasets from non-cardiac and cardiac surgery. Training is done on a part of the non-cardiac surgery dataset, and tests are performed separately on the rest of the non-cardiac dataset and cardiac dataset. Compared to a realistic clinical baseline, the proposed method achieves a significant improvement on the non-cardiac surgeries (precision of 42% compared to 28% for a recall of 24%). For cardiac surgery, this improvement is less significant but still demonstrate the generalization of the model.

Accurate forecasting of tropical cyclones is crucial for coastal regions to reduce the loss of life and property. Nearreal-time analysis data, such as the Cross-Calibrated Multi-Platform Ocean Surface Wind Vector (CCMP), can serve as guidance for tropical cyclone forecasting. Its gap-free coverage with uniform temporal and spatial resolutions is particularly useful for studies on air-sea interactions and ocean dynamics. However, its spatial resolution is 25 kilometers and is prone to significant underestimation of wind speeds for tropical cyclones due to rain contamination. Conversely, Synthetic Aperture Radar (SAR) can provide all-weather ocean surface wind measurements with a resolution of approximately 0.5 kilometers, but its temporal and spatial coverage is limited. We develop a novel deep learningbased method to fuse CCMP and SAR-derived ocean surface winds, integrating the strengths of both datasets. We utilize SARderived OSW associated with tropical cyclones worldwide as ground truth and incorporate corresponding CCMP as input to train the deep learning network to statistically downscale CCMP. We have evaluated various deep learning architectures, including U-Net, DeepLabV3+, and TransUNet. Our results show that TransUNet achieves superior performance, improving CCMP by 45% in root mean square deviation (RMSD) for wind speeds above 20 m/s, 20% in overall RMSD, 56% in maximum wind speed, and 65% in the radius of maximum wind for tropical cyclones. Our method produces uniformly distributed, gap-free, high-accuracy, and fine-resolution ocean surface wind data for tropical cyclones and has wide applications for downscaling spatiotemporal data.

Non-terrestrial networks (NTNs) contribute significantly to offering connectivity services for the upcoming sixth generation (6G). However, many services demand real-time uplink data transmission. Owing to the dynamic topology, instability of NTN channel status, resource constraints, and tremendous mobile users on ground networks; resilient and effective resource management and channel accessibility pose substantial challenges. This work investigates the uplink sum-rate maximization and latency minimization problems with a channel allocation scheme for user equipments (UEs) to high-altitude platforms (HAPs) of a particular NTN when UEs simultaneously transmit data to HAPs and HAPs forward it to low-earth orbit satellite (LEO-sat) via multiple subcarriers. We formulate a fairness resource allocation optimization problem reflecting the system utility and load of the HAP-LEOsat link for aerial-to-space uplinks. To tackle the high-dimensional non-convex optimization problems and short visibility of LEO-sat in the learning, we advocate a novel deep reward-based multi-agent reinforcement learning (DeepRMARL) with prioritized experience replay (PER) and priority sampling approach (PSA). Furthermore, we present a long-short-term memory (LSTM) model for future environment channel prediction due to time-varying channels between UEs and HAPs. The results explicitly unveil the feasibility of our method in optimizing total uplink rate, achieving a higher fairness index, minimizing latency by approx. 12.2%, and predicting future environment states with approx. 92.8% accuracy.

This paper presents a hybrid classifier that leverages the strengths of decision trees and genetic algorithms to improve classification accuracy. The classifier was implemented in Java using object-oriented design principles, resulting in a modular and maintainable code structure. The system consists of four main packages: data management, decision tree construction, genetic algorithm operations, and a graphical user interface. The genetic algorithm optimizes feature selection, which enhances the performance of the decision tree classifier by focusing on relevant features and eliminating noise. Results from experiments showed that the hybrid technique outperformed manual feature selection in three classification tasks: wine quality evaluation, water potability determination, and horse blood sample categorization. The hybrid algorithm's increased accuracy compensates the extra computing expense, despite its longer execution time. To improve feature evaluation, future research will examine new feedback mechanisms from the ID3 algorithm, optimize neural networks, integrate parallel computing, and conduct additional testing.

The Location Service ETSI standard has evolved to closing a critical gap between the standards for positioning mechanism of 5G mobile networks and the Multi-access Edge Computing (MEC) standard framework. Location Service (LS) generally allows, among other functions, to optimize services according to users' geographical location. Although LS already offers features typical of the next generation of mobile networks, such as 3D positioning of user equipment, the LS standardization framework is still evolving to meet the demands in support of next-generation mobile networks, most notably the sixth generation paradigm. Some of the features in the current standards, for instance, the concepts of zones, carry the promise of possibly transformative LS applications in future mobile networks. For instance, LS can help the building of a "6G network of (sub-)networks", with various technology solutions, towards the fully decentralized network operation. These and the related developments are driving further evolution of MEC and LS standards to support such new features. This article addresses the evolution of current Location API standards towards 6G mobile networks. We propose extensions and improvements to the current Location API data model to include data fields to characterize novel features and the envisioned novel use cases in the next-generation mobile networks. We show that the proposed modifications allow us to abstract the user's location into a tuple of physical and network locations, allowing MEC applications to perform complex data flow optimization.

The palpation of arteries has physiological significance. Existing pulse actuator designs intended to replicate the haptic cues of palpation focus primarily on the normal force interactions of the pulse, neglecting the shear forces arising from oscillations of the arterial wall and blood flow. This work aims to assess their magnitudes, facilitating the design of a high-fidelity sensor-actuator system for pulse palpation using simulations and experiments. The simulation yields the magnitudes of force and displacement of the wall in longitudinal, transverse, and radial orientations during palpation. A two-dimensional mathematical model of palpation computes the range of force interaction across different arterial elasticities. The experimental configuration featuring a hydro-electromagnetic actuator and a haptic sensor, termed Subblescope, validates the range of arterial force feedback and wall oscillation on soft and hard arteries. Validation also includes measurement of the pulse using the Subblescope of the radial artery of 18 subjects. The mathematical analysis provides the operating range of the sensor-actuator system as 0.005 N to 2.5 N. The reactive force along the Z-axis fluctuates between 10 mN to 300 mN, the transverse and longitudinal force along the Y and X axes ranges 6.25 mN to 18.75 mN and 1 mN to 20 mN respectively. The hard artery demonstrates a higher transverse oscillation compared to the soft artery by 20 µm. Radial displacement by indentation varies between 0 µm to 550 µm, as indicated by mathematical analysis, simulation, and experiment. The displacement along transverse and longitudinal axes ranges between 5 µm to 100 µm. The fluid-based actuation is necessary to create the haptic modalities of pulse under palpation. The hydroelectromagnetic actuator generates both normal force and shear deformation during pulsatile flow. Future research will explore creating training modules replicating pulse-haptics linked to various physiological states.

Optical telescopes are highly effective in tracking space objects and are indispensable for the space situational awareness ecosystem. However, due to their large focal lengths, typical telescopes only allow instantaneous observation within a very narrow field of view, which requires prior knowledge of the orbit and reliable tracking mechanisms. This paper presents a framework for blind detection and estimation of orbital parameters without prior knowledge of ephemeris, specifically when using wide field-of-view optical sensors. The framework is divided into three modules: (i) calibration of the imaging sensors to enable accurate mapping to the equatorial coordinate system, (ii) a differential processing algorithm for captured images to extract satellite trails from the celestial background, and (iii) orbital estimation based on iterative optimization, with a matchedfiltering concept at its core. Extensive night-sky modeling and emulation are employed as controlled validations of the framework to provide quantitative performance metrics. Additionally, real-life measurements were conducted using the state-of-the-art wide-field observatory MASCARA (Multi-site All-Sky CAmeRA), originally designed for exoplanet detection. The results of the experiment suggest the potential of the proposed method for space situational awareness using wide-field optical sensors.

This paper presents the design and implementation of novel dual-channel (MIMO hybrid) beamforming wireless front-ends operating in the D-band (sub-THz range), optimized within the 150 to 170 GHz frequency spectrum. The development incorporates analog beamforming networks realized through two 16x16 microstrip Rotman microwave lenses and dual sets of 1to-16 RF MMIC switches, in conjunction with a 16x2 antenna array comprised of low-profile on-silicon patch antenna elements and a PTFE elliptically extruded dielectric lens. The resulting architecture produces 32 distinct RF beams bifurcated into two sets of 16 beams, each oriented at two elevation angles. Each set is characterized by beams spaced 4 degrees apart in azimuth, collectively resulting in a 64-degree azimuthal scanning capability. Furthermore, the transmit front end is outfitted with 32 parallelized power amplifiers distributed across 8 InGaAs MMIC chips, equipping it to yield approximately 50 dBm effective isotropic radiated power (EIRP). In parallel, the receive front end comprises 32 low-noise amplifiers (LNAs) similarly arrayed across 8 InGaAs MMIC chips, each delivering an active gain exceeding 30 dB, which augments the overall passive antenna array gain of 25 dBi.

This research paper explores the application of game theory to analyze the welfare and competitive dynamics of Hindu temples operating in an oligopolistic environment. The problem centres on how temples, positioned as players in an oligopoly, can strategically choose between modernization and maintaining traditional practices to optimize social welfare and resource allocation. The methodology involves developing a game-theoretical model, defining two temples (Temple A and Temple B) as players with two strategy options: modernization or traditionalism. A payoff matrix was constructed to quantify the outcomes of each strategy combination, and Nash equilibrium was identified to determine the optimal strategy. The results reveal that while modernization yields mutual payoffs of 5 for both temples, sticking to tradition provides a higher combined payoff of 12, presenting a classic prisoner’s dilemma. The findings suggest that while modernization offers benefits, cooperation and a focus on traditional practices can result in greater overall welfare for both temples. This study demonstrates the usefulness of game theory in understanding economic and strategic decisions within religious institutions and highlights the importance of cooperative strategies in maximizing social welfare.

The convergence of contemporary automotive technologies and intelligent transportation frameworks has created significant opportunities for the application of predictive analytics and machine learning to refine vehicle performance and bolster driver security. In this endeavour, we harness the processing power of the Arduino Uno microcontroller, integrated with sophisticated machine learning methodologies, to deliver instantaneous tracking of fuel expenditure and nuanced categorization of driving styles. Specifically, our system employs an Arduino Uno board, augmented by an accelerometer and GPS module, to capture essential vehicular dynamic and positional data. The initial sensor readings are subjected to meticulous preprocessing, encompassing signal filtering and data partitioning, to guarantee their dependability for subsequent machine learning analysis. The extraction of key features, such as the magnitude of acceleration and fluctuations in velocity, further refines the precision of our predictive models. For real-time analytical purposes, we utilize Convolutional Neural Networks (CNNs) to discern driving patterns (e.g., forceful, balanced, economical) from temporal sensor data, and AdaBoost to estimate fuel usage based on derived attributes. Immediate feedback, displayed on an LCD screen, furnishes drivers with prompt insights and recommendations aimed at fostering fuel-efficient and secure driving habits. This project illustrates the transformative capacity of AI-driven intelligent transportation systems to encourage environmentally responsible driving practices and enhance road safety.

Cloud data centers serve as the backbone of modern digital infrastructure, supporting a vast array of applications from enterprise computing to artificial intelligence-driven services. However, ensuring fault tolerance while maintaining energy efficiency remains a significant challenge due to the high computational demand and power consumption of conventional cloud architectures. Reversible computing has emerged as a promising paradigm, offering ultra-low-power execution while minimizing heat dissipation, making it particularly attractive for next-generation cloud data centers. This paper explores the integration of fault-tolerant reversible computing architectures to improve cloud infrastructure security, reliability, and energy efficiency. Key strategies include leveraging reversible logic circuits for low-power data processing, designing fault-tolerant self-healing models for cloud resilience, and incorporating quantum and AI-driven techniques to enhance security and predictive fault management. In addition, the study investigates the role of reversible encryption schemes in mitigating security vulnerabilities while maintaining computational efficiency. A comprehensive analysis highlights the advantages, challenges, and future research directions of adopting reversible computing for secure and energy-efficient cloud data centers.